Wan 2.2 · Text to Video

This is the plainest video node in the Wan corner of the pack: a prompt in, one silent five-second clip out, 16:9 or 9:16, 480p or 720p, $0.24 a run. No input image, no audio, no LoRA slots, no knob for length. If you want to know whether your prompt idea works before you spend on something better, this is the cheap way to find out.

Where it sits

Wan 2.2 is the last open Wan - Alibaba moved 2.5, 2.6 and 2.7 behind an API and stayed there - which makes it the well-understood member of the family. Locally it's a 27B MoE with 14B active, a two-expert split between a high-noise pass (composition, motion) and a low-noise pass (detail), and it needs real VRAM management to run at all. Here, none of that is your problem: the node is an HTTP client, and the model runs on someone else's card.

What you give up is length and sound. Five seconds, silent, no duration widget. Nothing wrong with that for a shot test or a B-roll draft; just don't expect the 15-second clips the 2.6 nodes do.

The inputs

prompt is the whole job - subject, action, camera, light, pacing. Wan is unusually good at understanding movement, which is the reason people still put up with it, and the same trait means a prompt that describes motion gets better results than one that describes a photograph. "Slow dolly left as she turns toward the window, coat hem lifting, warm backlight" beats "a beautiful woman, cinematic, 8k".

aspect_ratio is 16:9 or 9:16 - vertical output is a first-class option here, which matters more than it sounds, since half the reason people reach for a hosted model is social-format output without a re-crop.

resolution is 480p or 720p. 480p is noticeably cheaper to iterate on and, at five seconds, good enough to judge whether the motion works. Once the take is right, regenerate at 720p.

enable_prompt_expansion is off by default. Turn it on when your prompt is a loose idea and you want the service to fill it out; leave it off when the words matter - a specific camera move or an unusual wardrobe detail - because expansion can rewrite the interesting part out.

seed maps to the same draw every time. That's the dial for A/B testing prompt wording without confounding it with randomness.

Wiring is trivial: video → Save Video. task id is a string for your console logs or for support if a run goes sideways.

Install

cd ComfyUI/custom_nodes
git clone https://github.com/spicyapi-ai/comfyui-spicyapi

Then restart and paste a key from spicyapi.ai/console/keys into Settings > SpicyAPI. ComfyUI Manager users can search "SpicyAPI" instead. There's nothing to compile and nothing to download - the pack's dependency list is empty because aiohttp, torch and Pillow all ship with ComfyUI already. No GPU, no VRAM, nothing to evict between runs.

The key is intentionally not a node field, so it never lands in a workflow file or the metadata of images ComfyUI writes. For an unattended machine, set SPICY_API_KEY in the environment and skip the settings panel entirely.

Getting burned, cheaply

This is the cheapest video node in the pack, which makes it the right place to learn the pack's cost habits. Every run is billed and there is no local cache to save you. Seeds default to fixed, so queueing the same graph twice charges once - leave control after generate alone until you actually want variety. Set Max cost per run in the SpicyAPI settings; anything quoted above it stops before the charge.

Cancelling is not a refund button. The job continues on SpicyAPI, is charged if it succeeds, and its output stays in your console for up to 14 days - so if you cancel, go look at the result rather than re-running and paying twice.

Error messages are blessedly specific. No SpicyAPI key is set means no key. 40004 means the option combination you picked has no deployment - change the parameter named in the message. 40310 means your account hasn't verified its email yet. 50302 means the generation failed upstream and was refunded, so re-run. And if the node list is missing models you saw on spicyapi.ai, restart ComfyUI: the list is rebuilt from your account catalogue on startup.